Localizely MCP server

The Model Context Protocol (MCP) is how an AI assistant reaches the tools you already use. Connect Localizely and your assistant can add the string keys of the screen you just wrote, read the German translation that is breaking your layout, or tell you what is still untranslated before you cut a release, without you switching to the browser.

It is hosted, so there is nothing to install. The server reads your localizely.yml through the assistant, which means the project, the branch and the languages are already known.

Setup

Use the API token from your account page — the same one the CLI uses.

Things to ask for

Once it is connected, you talk to your assistant rather than to the tools. For example:

• “Pull the hardcoded strings out of CheckoutScreen.dart into Localizely and translate them to German”

• “What is still untranslated in Spanish on the release branch?”

• “The German checkout button is too long, shorten it in Localizely”

• “Rename btn_ok to common_action_ok everywhere, including Localizely”

• “Which key is the text “You have pushed the button this many times”?”

Tools

Translation files

Uploading and downloading translation files is not a tool here, and that is deliberate: a hosted server cannot reach your filesystem, so the bytes would have to travel through the assistant. The CLI reads and writes them directly instead, and your assistant can run it. The MCP server tells it so, with the commands below, so in practice you just ask for the files and it runs them.

Both commands take their project, file paths and formats from localizely.yml, so there is nothing to pass. If the CLI is not installed yet, your assistant can fetch the binary for your platform from the GitHub releases, or run it through Docker without installing anything. See the CLI page for every installation option.

Tokens and permissions

• A read-only token exposes only the tools that read, so you can hand one to an assistant you do not want writing to your project.

• The token belongs to a user, and that user needs the owner or admin role on the project. A token from a translator or a reviewer is refused.

• API access, and so the MCP server, is available on paid account plans.

• To revoke an assistant's access, delete its API token on your account page.

While it is in beta

The tools and their inputs may still change while we learn how assistants use them, so pin down anything you automate. Tell us what is missing and it will shape what comes next.